Andrew Nichols

Generally this is how we think about incentives…

Why is this a problem?

the implications

Imagine policing as an example.

It was originally thought that you could model somones desire to engage in crime as a utilitiy function, and that, if you simply levied up the costs associated with engaging in a crime, you could incentivze a selfish optimizer to not commit a crime.

And here’s how that worked out

Contrast this with commmunity policing models, which incorporates the community and assumes social incentives to not engage in crime (this guy is passing out food and water to neighborhood kids).

so what’s an alternative?

My experiment is designed to explore these social preferences, but i’ll need to expand on the theory just a bit

the theory

How would we represent these social preferences in a utilitiy function?

and mapped back on to the diagram…

the experiment

Me: “Can you help me complete this survey?”

stranger: “Yeah sure.”

Me: “I’ll give you a quarter to help me complete this survey”

Stranger: “No thank you”

I’ll be adminstering a fake survey, in which I offer no compensation, “categorical” compensation, or high compensation. The prediction is that there will be almost a stepwise dropoff at very low values of compensation as compare to offering no compensation, but gradually the perceived material incentives will increase the frequency as the amounts get larger.

So how will my model test for this?

the model

The model is sepcified like this:

\(probability(help) = \beta_0 + \beta_1offer + \beta_2offer^2 + \epsilon\)

Which is to say that i’ll be looking for a quadratic shape where the frequencies for no compenstation and high compenstation are greater than the frequencies for categorical compensation.
I would broadly expect…

\(\beta_1 < 0\) (the general downward drop representing the categorical effect)
\(\beta_2 > 0\) (the bend up representing the gradual dominance of the material incentives)

the data

The data will be relatively simple, either yes or no, and a series of controls. The control questions will actually be the content of the “survey”. The ones I preliminarily intend to include are…
`

frequency - a one if they offered to help
female - a one if they appeared to be female
age - how old they were
ethinicity - what race/ethincity were they

key challenge will be dealing with the self selection (ie the people who turn me down, how do I get control data on them)

the inspiration

questions / feedback?